Evaluation of Interference Cancellation Architectures for Heterogeneous Cellular Networks

Contents

Notations

Acronyms

1 Introduction
1.1 Motivation and Objective
1.2 Outline and Organization of the Thesis

2 Fundamentals
2.1 Bit-Interleaved Coded Modulation
2.1.1 System and Signal Model
2.1.2 Low Complexity LLR Metrics for BICM Receivers
2.2 Interference-Aware System
2.2.1 IA Receiver: System & Signal Model

3 Structure of the Simulator
3.1 System Parameters & General Code Structure
3.2 Radio Propagation Channel
3.2.1 Simulation of AWGN-Channel model
3.2.2 Path Loss Channel Model
3.2.3 Simulation of Rayleigh Fading Channel
3.3 The Baseband Part of the Transmitter
3.3.1 Convolutional Coding and Puncturing
3.3.2 Bit-Interleaver
3.3.3 Bit-Level Scrambling
3.4 Interference Model
3.5 The Baseband Part of the Receiver
3.5.1 Pilot-based Channel Estimation
3.5.2 De-Puncturing and Soft Output Viterbi Decoding
3.5.3 Metric Computing Device
3.6 Base Stations Channel Estimation Enhancement
3.6.1 Serving Base Station: Holes
Contents
3.6.2 Interfering BS: Pilot Boosting

4 Simulation Results

5 Summary and Outlook

Bibliography

List of Figures

2.1 Block diagram of BICM transmission: encoder ENC, bit interleaver n, modulator M, metric computing device M-[1], deinterleaver n-[1] and de­coder DEC

2.2 Binary labelling sets for QPSK and 16-QAM

2.3 Co-channel interference between hexagonal cells with a frequency reuse of factor one

3.1 Structure of the Simulation Testbed

3.2 Illustration of one LTE Slot

3.3 Power Spectrum of LTE DL Signal

3.4 AWGN Channel model

3.5 generated AWGN, where z e CN(0,0.35)

3.6 AWGN-Channel model: Impact on one OFDM symbol

3.7 BER theoretical vs simulated

3.8 Rayleigh Channel Model

3.9 Non Line of Sight (NLOS)- Multipath

3.10 Rayleigh Distribution and Example

3.11 BER performance in a Rayleigh Propagation Channel

3.12 Block Diagram of the Baseband Transmitter

3.13 OFDM Modulation, OFDM Symbol

3.14 Encoder with generator polynomials g1 = [111] and g2 = [101]

3.15 Trellis diagram of (7, 5)8 encoder

3.16 Interference Model

3.17 Baseband Part of the Receiver

3.18 Pilots within a Subframe

3.19 Channel Estimate of a 16-taps Rayleigh Channel by SNR = 15 dB

3.20 2 x 1-D Interpolation/Extrapolation

3.21 Linear Interpolation in the Frequency-Domain

3.22 Metric Computing Device: Metric as Log Likelihood Ratio

3.23 LLR vs Bit

3.24 1tap Rayleigh Channel effects on an OFDM Symbol by SNR = 5 dB . . .

3.25 OFDM symbol after usual Phase and Amplitude Equalization

3.26 OFDM symbol after Matched Filter

List of Figures

3.27 H1 vs. SIR

3.28 H2 vs. SIR

3.29 IAR Performance Boundaries

3.30 Serving BS enhancement: Holes at the pilot positions of the interfering signal

3.31 Interfering BS: Pilot Boosting

4.1 BER performance of IA-R vs. II-R by SNR = 5 dB and variable SIR and Perfect Channel Knowledge

4.2 BER performance of IA-R vs. II-R by SNR = 20 dB, variable SIR and Perfect Channel Knowledge

4.3 IA-R vs. II-R by Perfect Channel Knowledge

4.4 BER performance of IA-R vs. II-R by SNR = 15 dB and variable SIR and LS Channel Estimation

4.5 IA-R vs. II-R by LS Channel Estimation

4.6 IA-R Performance with Base Station Interference Cancellation Enhance­ment, by SNR = 5 dB and variable SIR

4.7 IA-R Performance with Base Station Interference Cancellation Enhance­ment, by SNR = 20 dB and variable SIR

4.8 Proposed Interference Cancellation Architectures performance by variable SNR/SIR values

Notations

Abbildung in dieser Leseprobe nicht enthalten

Acronyms

Abbildung in dieser Leseprobe nicht enthalten

1 Introduction

1.1 Motivation and Objective

With the increasing data throughput requirements, the cellular network needs to move from homogeneous to heterogeneous system. In fact, the coexistence of different types of base stations with different capabilities such as femto/pico base stations as well as relays and macro base stations in random placements should improve the coverage and the spectral efficiency of the cellular networks.

However, the complexity of inter-cell interference management will grow drastically and traditional interference avoidance/mitigation approaches need to be revised. Approaching this problem at the user equipment (UE), is of great interest since it can rely on little coordination among base stations.

The work presented in this thesis focuses on a downlink interference cancellation at the UE and shows that such an intelligent receiver can bring its promised benefit only if the base stations get involved in the interference cancellation, specifically in the channel estimation process. The limitations of this approach are evaluated and depending on the surrounding base stations two solutions are proposed and discussed.

1.2 Outline and Organization of the Thesis

This thesis consists of four chapters, that are organized as follows:

Chapter 2 first introduces the basic concepts for the understanding of the thesis. It defines the bit-interleaved coded modulation (BICM), the low complexity log-likelihood (LLR) metrics for BICM receiver and the interference-aware receiver (IA-R).

In Chapter 3, the structure of the simulator is described. This includes transceiver sys­tem model as well as the propagation channel and the interference model. Furthermore, two solutions for IA-R performance enhancement are proposed.

Chapter 4 contains the simulation results of the considered systems. First, a compari­son between the IA-R and the interference-ignorant receiver II-R under perfect channel knowledge are evaluated, providing an estimate of the upper performance bound of the IA-R. Then, the performance of the IA-R under more realistic assumptions, with and without base station interference cancellation enhancement, is tested.

Finally, Chapter 5 summarizes the thesis and gives a brief outlook on future research and developments.

2 Fundamentals

2.1 Bit-Interleaved Coded Modulation

2.1.1 System and Signal Model

Bit-interleaved coded modulation (BICM) is the serial concatenation of a channel en­coder, a bit-wise interleaver n and an M-ary modulator, which maps blocks of m bits to a QAM constellation x from a symbol set X C CN of size |X| = M = 2m.

Abbildung in dieser Leseprobe nicht enthalten

Figure 2.1: Block diagram of BICM transmission: encoder ENC, bit interleaver n, modulator M, metric computing device M-[1], deinterleaver n-[1] and de­coder DEC

As shown in Fig. 2.1 the information bits are first encoded by a convolutional en-coder and then a pseudo-random bit interleaver permutes the time index i of the coded bits in order to decorrelate the bits associated with the given symbol and to disperse the burst errors. Every m coded and interleaved bits are then mapped into a M-ary QAM symbol. It is possible to employ any two-dimensional constellation but in [2] was shown that for Gray encoded BICM systems, the calculation complexity for each bit Log Likelihood Ratios (LLR) can be drastically reduced without compromising the system performance [12].

Gray mapping was first proposed by Frank Gray and is defined in [4] as follows:

Abbildung in dieser Leseprobe nicht enthalten

In other words, it means that in Gray mapping, adjacent constellation points differ only in a single bit (Fig. 2.2), which reduces the mean square error by one bit error.

Figure 2.2: Binary labelling sets for QPSK and 16-QAM

Abbildung in dieser Leseprobe nicht enthalten

On the receiver side, there are a metric computing device M-[1], de-interleaver n-[1] and a soft decision Viterbi convolutional decoder, as suggested by Zehavi in [23].

The main task of the outer^[1] receiver is to find the most probable code bit uU for each bit position i, i = 1,...,m, within the k-th frequency tone, given a received signal vector y, which can be obtained by

Abbildung in dieser Leseprobe nicht enthalten

This detection rule is referred to as maximum a posteriori probability (MAP) criterion and it gives the optimal reliability information for a certain bit position i of the received vector y.

Since the outer receiver has to take in account the memoryless binary propagation chan­nel H, which is characterized by an output alphabet Y and the conditional probability^[2]p(y\ui = b), y eY, a maximum likelihood (ML^[3]) decoder would be more appropriate in this case, which can be straightforward obtained from (2.1) by using Bayes’ theorem. Now, assuming P(b) to be equal for all possible values of b, which is a very fair assump­tion, since the transmitted bits are randomized by means of a a pseudo-random scrambler and interleaver (see [21] and [15]), MAP and ML criterion become equivalent:

Abbildung in dieser Leseprobe nicht enthalten

2.1.2 Low Complexity LLR Metrics for BICM Receivers

In this section, a low complexity LLR metric calculation is derived when optimum (un­quantized) soft decoding at the receiver is employed. This type of receiver is also referred as interference ignorant receiver.

Let us consider the simplest possible scenario, which is a random BPSK symbol trans­mission over an AWGN channel. As we know, BPSK modulation can be defined as x = (-1)u, where u e {0,1} and x e {±1}.

Thus, the probability distribution function for the received symbol y is given by

Abbildung in dieser Leseprobe nicht enthalten

Instead of probability and for simplicity’s sake we can use the ML metric \(y, x) = max lnp(y\x)

Abbildung in dieser Leseprobe nicht enthalten

Actually, ML metric involves summation instead of maximum operation and the above metric is commonly referred as max log MAP metric [12].

So, max log MAP metric is equivalent to pick up the symbol x with the least Euclidean distance to the received symbol y, D(y,x) = \ \y — x||[2]:

Abbildung in dieser Leseprobe nicht enthalten

Since the decision is a comparison problem between two possible log-likelihoods ln p(y \x = +1) and lnp(y\x = —1), a ratio would be more suitable for the solution

Abbildung in dieser Leseprobe nicht enthalten

where A(y, x) e R is the so-called log-likelihood ratio.

The sign of A(y, x) corresponds to its hard decided binary value. A positive LLR in­dicates that the bit is more likely to be 0, a negative LLR indicates that the bit X is more likely to be 1 and the magnitude A(y, x) indicates how sure we are about the hard decision of the decoded bit. But the decision will be first taken in the soft Viterbi decoder. Thus, the data is passed from one component to another in form of LLRs. Clearly, practical digital implementations can only use a fixed point approximations of the real numbers (LLR quantization), which is obviously not a trivial task [10, 9] and it is out of the scope of this thesis.

Obviously, the BPSK example was so simple that is not possible to talk about a simplified LLR. So, let us consider a higher modulation scheme such as a Gray coded QPSK constellation (Fig. 2.2a), where x[k] = xi[k] + jxQ[k] denote the transmitted symbol on the k-th subcarrier over a flat-fading real valued channel. The received symbol is then yk hk xk + zk.

So, formulating the max log MAP metric from (2.4) in terms of real and imaginary parts and taking the channel into consideration, we obtain:

Abbildung in dieser Leseprobe nicht enthalten

where yk = h^yk is the matched filter output, the subscripts (-)R and (■)i indicate the real and imaginary part of the complex symbol, respectively and since |y|[2] is common for all A, it can be omitted in the computation.

Using (2.7) in (2.6), we obtain:

Abbildung in dieser Leseprobe nicht enthalten

Furthermore, we can decouple the real and imaginary parts of (2.8):

Abbildung in dieser Leseprobe nicht enthalten

From Fig. 2.2a we can also see that when the bit x\ in the QPSK constellation toggles from 1 ^ 0 only the real part of the constellation is affected, and for the bit x2 only the imaginary part. And if we look further, in the 16-QAM or even higher Gray mapped modulation schemes we conclude that

From (2.10), the LLR of the first bit for QPSK with a normalization factor of — is given as

Abbildung in dieser Leseprobe nicht enthalten

and the LLR of the second bit is given as

Abbildung in dieser Leseprobe nicht enthalten

Most of the simplifications occur just with replacing {xRj} with its actual value {±}.

2.2 Interference-Aware System

In the last section a simple and general expression of max log MAP metric computation in a flat fading channel, without any co-channel interference was introduced. In this section we will go further and consider a more realistic scenario, which takes into account co-channel interferences between hexagonal cells. Co-channel interference occurs when, a MS simultaneously receives signals from the serving eNodeB , as well as from co­channel neighbour eNodeB. Indeed, the downlink (DL) performance of a cellular system is strongly limited by co-channel interferences, especially in the cellular systems with a frequency reuse of factor one or fractional frequency reuse systems [11].

Abbildung in dieser Leseprobe nicht enthalten

In most cases, there are a maximum of two dominant interferers, one if MS is close to the cell boundaries, two near cell edges. For simplicity sake and without lose of generality, only one interferer in this work is taken in account.

2.2.1 IA Receiver: System & Signal Model

We consider a single frequency reuse cellular network, with two neighbour cells, which use a BICM based OFDM system for DL transmission and one IA-receiver, as shown in Fig. 2.3 One signal of interest and one interferer. We also assume that the CP is of appropriate length, the BS’s are synchronized for transmissions, the IA-R knows about the modulation scheme of the interferer and can also estimate its propagation channel.

Cascading the IFFT at the BS and the FFT at the MS with CP extension, transmission at the k — th subcarrier can be then expressed as follow:

Abbildung in dieser Leseprobe nicht enthalten

where K is the total number of subcarriers, Hk — [hi;k h2,k] is the virtual channel from two BSs to the user at the k — th frequency tone, xk — [x\,k x2,k]T and zk is the noise vector with zk e CN(0,a[2]). Each subcarrier corresponds to a symbol from a constellation map x\ e Xi and x2 eX2.

The max-log MAP bit metric for bit b of the signal of interest x\ can be then obtained by inserting the new received signal model yk from (2.13) in (2.6)

Abbildung in dieser Leseprobe nicht enthalten

where yi — hHy, and y2 — hHy are the output of the MF and p12 — hHh2 is the correlation coefficient between the two channels. Splitting (2.14) into real and imaginary parts and omitting the common term ||y||[2] we have

Abbildung in dieser Leseprobe nicht enthalten

Thus, if |Xi| — M and |X21 — M', we need to compare MM' possible values and take the minimum between them, which can be by high modulation schemes very inefficient in terms of performance and computational complexity.

Lets define

Abbildung in dieser Leseprobe nicht enthalten

(2.15) can be minimized if x2r and x2,I are in the opposite directions (have the opposite signs) of ni and n2, respectively.

Let X1 and X2 be each a normalized symbol set with a Gray mapped QPSK constellation. So, |x1|[2] = |x2|[2] = 1 and since (||h1||[2] + ||h2||[2]) and the factor 2 are common for both max log MAP metrics X' with i = 1,2 , they can be neglected.

Hence, (2.15) can be written as

Abbildung in dieser Leseprobe nicht enthalten

Abbildung in dieser Leseprobe nicht enthalten

From (2.17), the LLR for the first bit of the QPSK symbol of interest x1 is given by

Abbildung in dieser Leseprobe nicht enthalten

and the second bit of the same QPSK symbol is

3 Structure of the Simulator

This chapter is dedicated to the Matlab implementation of a BICM based OFDMA transceiver system in the equivalent baseband. We assume that the time and frequency resources are organized as in the LTE standard, but without considering all the details and only the downlink transmission scheme is considered.

Each implemented block will be explained and discussed. A particular focus here is the performance of the interference aware receiver in an inter-cell interference environment with one dominant interferer. Additional optimization techniques are also proposed and discussed in this chapter such as boosted pilot symbols in the interferer and REs puncturing in the signal of interest.

In fact, LTE standard allows a certain degree of freedom in the signal generation, which makes these two optimization techniques easy to incorporate in real systems.

3.1 System Parameters & General Code Structure

Fig. 3.1 shows the general code structure of the simulation testbed, which is composed of five main building blocks:

- Control unit, which is used to control the behaviour and priorities of the different components of the implementation, some of the parameters are set as default and others are adjustable.
- Serving eNodeB,
- Interferer eNodeB,
- Virtual channel,
- and the UE

Each of these block will be discussed in the following sections, where the interferer eNodeB and the serving eNodeB are both summarized in the transmitter section. The major difference between these two kind of transmitters is notable only if one of the optimization techniques introduced above is considered.

In (3.1), the essential system parameters of a 5MHz DL PHY are illustrated.

Abbildung in dieser Leseprobe nicht enthalten

Figure 3.1: Structure of the Simulation Testbed

LTE transmissions are organized into 10ms radio frames, which corresponds to 76800 samples by Nftt = 512. Each of these radio frames is divided into 10 equally sized subframes. Each subframe consists of 2 equally sized slots with Tsiot = 0.5ms. In the frequency domain, one slot consists of 7 OFDM symbols, where each OFDM symbol consists of a certain number of active subcarriers Nc. The REs of the first symbol of the implementation are filled only with pilot symbols, while the REs of the next two symbols are left empty, since the control channels in the implementation are not considered.

Fig. 3.2 illustrates one OFDM slot. In the time domain it consists of 3840 samples with Ts = 130.21ns and in the resource grid illustration it consists of 25 RBs x 7 Symbols. From the illustration of the power spectrum of the generated LTE signal, we can also see the 5MHz bandwidth as well as the sampling rate.

In the simulation testbed only one slot will be generated, sent and processed in the context of a Monte Carlo simulation.

Abbildung in dieser Leseprobe nicht enthalten

Figure 3.2: Illustration of one LTE Slot

Abbildung in dieser Leseprobe nicht enthalten

Figure 3.3: Power Spectrum of LTE DL Signal

[...]

---

^[1]A receiver can be divided in an inner and outer part. The transmission parameters such as carrier frequency offset, sampling clock offset, timing synchronization, I/Q-imbalance and multipath channel are estimated and compensated in the inner receiver. The compensated signal is then used to decode the transmitted data in the outer receiver, which mainly consists, in the case of a BICM based OFDM system, of a soft-bit computing device and a Viterbi decoder

^[2]Capital P(x) and lower-case p(x) are used for discrete probabilities and continuous probabilities, respectively

^[3]ML decoder maximizes Prob(received|transmitted). MAP decoder maximizes Prob(transmitted|received).